Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Case Study: Resilient Backbone Design for IPTV Services

      Meeyoung Cha∗ Gagan Choudhury† , Jennifer Yates† , Aman Sha...
ever, we consider the performance requirements to the ex-                           multiple Gb/s of aggregated capacity t...
from SHOs to all VHOs. This indicates that multicast can                         One of the challenges of using an integra...
service node is connected to a backbone node such that the                          can be optically protected or not – in...
3.2    Optical Based Network Designs                                      signed specifically for carrying only the IPTV tr...
Multicast 3Gbps + Unicast 0Gbps                                                                                           ...
Upcoming SlideShare
Loading in …5

Case Study: Resilient Backbone Design for IPTV Services


Published on

  • Be the first to comment

  • Be the first to like this

Case Study: Resilient Backbone Design for IPTV Services

  1. 1. Case Study: Resilient Backbone Design for IPTV Services Meeyoung Cha∗ Gagan Choudhury† , Jennifer Yates† , Aman Shaikh† , and Sue Moon∗ , ∗ † KAIST AT&T Labs – Research,,,, ABSTRACT tive pressures are providing network providers with the op- This paper presents a case study of designing resilient back- portunity to carry video on their terrestrial infrastructures [9]. bone network for supporting IPTV services within a single Today, these terrestrial networks typically equate to IP and network provider. We first introduce the architecture and optical backbones. the characteristics of IPTV traffic. We then explore the IPTV introduces a number of interesting challenges for the design space in terms of IP versus optical technologies, hub- research community. One fundamental question is: what is and-spoke versus meshed service layer topology, dual-homed the best network architecture for supporting IPTV? versus ring access, use of multicast, routing, and fast failure In this study, we limit ourselves to the problem of backbone restoration. From this design space, we propose a number network design, and do not focus on distribution out to- of design instances and evaluate them according to the cap- wards the customers. Even though IP forms the basis of the ital expense they incur. We demonstrate significant benefits architecture for most network providers, this does not nec- of multicast in reducing capital expense for broadcast TV, essarily translate into using an IP backbone for distributing illustrate that our particular switched optical network de- traffic over the wide area. An alternative is to use an optical sign requires less capital than the IP based designs, and infrastructure to distribute the traffic. Beyond technology show that ring access to the backbone is more attractive for decisions, network designers must also incorporate topology, broadcast TV, while dual-homed access being more attrac- routing, and performance considerations in designing a reli- tive if we have high volume of realtime Video on Demand able and cost effective distribution network. (VoD) traffic. In this paper, we provide what we believe to be the first detailed comparison of different architectural alternatives for Categories and Subject Descriptors supporting IPTV within a single network provider. We fo- cus on the capital expenditure associated with these designs. C.2.1 [Computer-Communication Networks]: Network Our work is not a complete exposition on IPTV service de- Architecture and Design; C.4 [Performance of Systems]: signs; nevertheless, we aim at grasping the essential high- [Design Studies, Reliability, availability, and serviceability] level design principles and trade-offs by rigorously exploring the design space. General Terms One purpose of this paper is to address the design space to Algorithms, Design, Management, Reliability support IPTV services in the backbone distribution network. These design choices are discussed in light of technologies, Keywords hierarchy, routing, and failure restoration. Another purpose Internet Protocol TV (IPTV), Network Provider, Network is to present how the design space can be realized in an op- Design, IP Network, Optical Network, Capital Expense erational backbone network. Especially, we focus on a few key architectural alternatives, namely: 1) integrating IPTV 1. INTRODUCTION AND MOTIVATION services with an existing IP based network; 2) constructing a dedicated overlay on top of an existing IP based network; The telecommunications market is rapidly evolving to of- 3) constructing a new point-to-point interconnected flat IP fer commercial-grade live broadcast TV and Video on De- network; and 4) integrating with an existing switched op- mand (VoD) over IP – known as Internet Protocol TV (IPTV) [9]. tical network. Accordingly, we propose a number of design World-wide, the number of IPTV subscribers was reported instances and evaluate them according to the capital expense to surpass 4 million in 2005. Scaling this to mass markets they incur. We also show the efficacy of using multicast over in today’s highly competitive environment necessitates an unicast for broadcast TV. extremely reliable and cost effective network infrastructure From a performance point of view, there is a tight inter- – all the way from the central head ends where the video is play between the application and the network design. The sourced, to the customers. application can be designed so as to best utilize the avail- Nation-wide delivery of broadcast TV to video head ends able network; similarly, the network must have adequate has traditionally been achieved via a satellite-based infras- performance to support the application. If the application tructure. However, satellite capacity exhaust and competi- is highly sensitive to loss (e.g., little or no buffering is em- ∗ These authors are supported by KOSEF through AITrc. ployed), then the network must have minimal loss and sup- port ultra-fast failure recovery. Discussing detailed require- ments of IPTV services is beyond the scope of this paper, To appear in the IPTV Workshop, International World Wide Web Confer- and we refer to the details of the ongoing work in [9]. How- ence, May 23, 2006, Edinburgh, Scotland, United Kingdom. 1
  2. 2. ever, we consider the performance requirements to the ex- multiple Gb/s of aggregated capacity toward each VHO. Al- tent that they impose specific fast failure recovery and avail- though this in itself is not a very large value, if we multiply ability constraints that impact our network designs. this by tens of VHOs, it becomes a massive amount of traffic. The rest of the paper is organized as follows. In §2, we Since identical content is to be distributed across all VHOs, introduce the overview of IPTV services. In §3, we present a we consider the use of multicast capabilities to minimize the case study of design space exploration and present a number communication cost (e.g., bandwidth consumed) [3]. of pragmatic design alternatives. Assessment of the designs VoD content is sent to each individual user as a real- follow in §4, and we conclude in §5. time dedicated stream. However, to minimize the amount of traffic that must be carried across the backbone, we as- 2. OVERVIEW OF IPTV SERVICES sume that popular VoD content is stored at the VHOs, and We start out by describing the service level architecture sourced from the VHO as and when requested by the end and traffic characteristics of IPTV services. users. When new content becomes available to the SHOs, it is pushed from the SHOs to the VHOs during off-peak pe- 2.1 Service Architecture riods. These transfers do not require realtime delivery, and The IPTV service architecture we envision includes a back- bulk-transfer applications (e.g., ftp) can be used to ensure bone and multiple regions [5]. We assume that there are reliable delivery. Since the traffic is assumed to be carried two locations from which IPTV traffic is sourced – these are during low utilization periods and does not require realtime known as the Super Hub Offices, or SHOs. Two SHOs pro- transmission, it has minimal impact on the network design vide redundancy to ensure reliable video transmission, even and architecture. We thus do not take non-realtime VoD in the face of catastrophic failure of one of the SHOs. In our traffic into account while designing the backbone network. study, we assume that both SHOs are always live to ensure If the service provider decides to offer a vast variety of rapid recovery in the event of a SHO failure. VoD content, it may not be cost effective to store the en- Video streams transmitted from the SHOs are received at tire content at every VHO. From an economic stand-point, the Video Hub Offices, or VHOs, where the video streams it generally makes more sense to store popular VoD con- may be further processed (e.g., advertisement insertion) and tent at the VHOs ahead of time, but source more esoteric then transmitted out towards the customers. VHOs also content from the SHOs in realtime, as and when customers store local video content to support VoD. Each VHO consists request it. Sophisticated cache management algorithms may of video equipment and two routers (for redundancy), where be used to increase the hit ratio of the requested VoD con- these routers interconnect the backbone with the regional tent. However, some percentage of the VoD content will network. For simplicity in this paper, we collectively refer have to be sourced from the SHOs to serve realtime cus- to the SHOs and VHOs as service nodes and refer to the tomer requests. We assume unicast delivery for those VoD routers within a service node as service routers. Figure 1 content, and take the bandwidth requirement of such traffic illustrates the service architecture. into account in our backbone design analysis. It is worth SHO pointing out that the realtime VoD downloads are expected Super Hub Office (SHO) to have greater traffic variability during peak usage periods compared with the broadcast TV traffic. Backbone Distribution Network 3. DESIGNING THE BACKBONE NETWORK VHO We focus on designing the backbone network intercon- Regional Regional Network Broadcast TV necting the service nodes. We start our discussion with the Network VoD VHO main axes of the design space and various options available along each axis. A set of design instances are realized by Video Hub Office (VHO) Regional Regional combining options available along each of these axes. Network Regional Regional Network The first axis deals with the technology used for inter- Network Network connecting the service nodes. We consider two alternatives: layer 1 (optical) and layer 3 (IP/MPLS) technologies. By optical network, we refer to an infrastructure consisting of Figure 1: Service architecture of IPTV optical components that provide fixed bandwidth “pipes” This paper focuses on the backbone part of the service ar- (or links) interconnecting the service routers. On the other chitecture, i.e., the network that connects SHOs and VHOs. hand, by IP network, we assume that traffic between the The design of the regional networks for connecting VHOs to service nodes is routed over intermediate IP routers. end hosts (or customers) is outside the scope of this paper. The second axis is the service layer topology. We con- sider two options: hub-and-spoke and meshed topologies. 2.2 Characteristics of IPTV Traffic In a hub-and-spoke topology, the SHOs are directly con- We consider three types of IPTV traffic being carried over nected to every VHO. In this topology, the service nodes the backbone network: broadcast TV (realtime), VoD down- only source/sink traffic. In a more highly meshed (or ring loads (non-realtime), and realtime VoD (realtime). Broad- based topologies), service nodes carry “through traffic” in cast TV traffic consists of traditional TV, High Definition addition to acting as sources/sinks of traffic. Thus, traffic TV (HDTV), and CD quality music channels. To be com- destined to each VHO may pass through multiple interme- petitive with today’s cable and satellite TV offerings, IPTV diate VHOs before reaching its destination. must offer hundreds of TV channels, which are typically 2 Finally, we consider the option of using multicast [3] ca- to 6 Mb/s or 6 to 12 Mb/s, depending on whether they are pabilities to reduce capacity required for broadcast TV. As standard or high definition TV signals. This translates into mentioned earlier, identical broadcast TV content is streamed 2
  3. 3. from SHOs to all VHOs. This indicates that multicast can One of the challenges of using an integrated network to be used to reduce the overall bandwidth consumed by such support both Internet and highly sensitive IPTV traffic is traffic. In this work, we also investigate how much value that the IPTV traffic is at risk of being impacted by the multicast capabilities provide us, compared to using unicast. Internet traffic (e.g., congestion, DOS attacks). Measures Having discussed the main axes for designing the back- should be taken to minimize such negative impact, for ex- bone, we next present how to combine various options to ample: provide adequate priority forwarding of the IPTV come up with design instances. Since the technology has traffic (using Diffserv [7], e.g.); implement measures to iso- the biggest impact on what other options we can choose late IPTV traffic from DOS attacks; and provide IPTV traf- and the resulting cost, we divide our discussions into IP and fic with preferential fast failure recovery (faster than that of optical based backbone designs. best effort traffic) where feasible. We assume that routing for unicast traffic is based on tra- 3.1 IP Based Network Designs ditional Interior Gateway Protocol (IGP) algorithms such as OSPF [10] or IS-IS [2]. These algorithms route traffic over We start by considering a network provider carrying IPTV the shortest path between source and destination, based on traffic over an IP infrastructure. At one extreme, we con- the administrative weights assigned to each network link. sider a dedicated IP network constructed purely to carry For multicast traffic, we assume that there is a single mul- IPTV traffic. This has the advantage in that the design can ticast tree including the two SHOs and all the VHOs. The be customized to support IPTV services, and that IPTV SHOs are assigned a common anycast source address so that traffic is not mixed with traditional Internet traffic, thereby each VHO is routed to the nearest SHO as it joins the mul- isolating this sensitive traffic from the perils of the pub- ticast tree. Thus, traffic is transmitted from the nearest lic Internet. The other extreme is to use a single common SHO to each VHO, and each VHO receives a single copy of network to carry all traffic – including Internet, VPN, and the multicast traffic at any given point in time. Should a IPTV services. Such an approach simplifies network man- SHO fail, then the VHO is automatically re-routed to the agement, in terms of having only a single network to design surviving SHO. The multicast routing is realized using the and operate, but requires careful performance management PIM-SSM (protocol independent multicast - source specific to ensure that the IPTV traffic receives high priority for- multicast) protocol [1, 6], which uses a shortest path multi- warding and is isolated from the vagaries of the Internet. cast distribution tree (MDT) rooted at the source. Receivers Intermediate solutions also exist, for example, overlaying a join the tree along the reverse shortest path based on the dedicated topology on top of an existing infrastructure. This sum of the administrative link weights. can be achieved by using the common backbone routers, but Rapid failure recovery is important in minimizing the neg- with separate links to carry IPTV traffic, providing a level ative impact of network outages longer than 50 to 100 mil- of isolation for the highly sensitive IPTV traffic. liseconds. Note that if we only use traditional IGP routing protocols and PIM-SSM, failure recovery may take up to a 3.1.1 Integration with an Existing IP Backbone few seconds. While such delay may be acceptable with many We assume a scenario in which a network provider al- applications, IPTV application requires well below 1 second ready has an existing IP network over which they will in- failure recovery in order for the end users to not feel the corporate the new IPTV demands. As mentioned before, impact at all. The IPTV application servers can also assist only the backbone links are shared and access links are ded- by incorporating buffers, so that the application can ride icated for IPTV traffic. Utilizing an existing infrastructure out short-term outages. Buffering necessitates delaying the enables rapid deployment of the new services, with minimal signal transmission, which in many broadcast applications overhead and efficient utilization of the network resources. has no effect on the customer’s satisfaction. However, for an IPTV traffic is special in that it is mostly uni-directional avid sportsperson, for example, delaying a live broadcast is (from SHOs to VHOs) and requires high bandwidth. More- simply not acceptable – especially if one tries to synchronize over, in the case of VoD, we expect high traffic variability across multiple broadcast mediums (e.g., radio and TV). during the peak usage hours. Utilizing a common infras- It is thus critical that the backbone network be able to tructure offers the potential to share bandwidth between rapidly recover from network outages. We consider two en- applications. For example, Saturday night may be a high hancements in providing fast recovery against link failures, load period for IPTV, as couples sit down to watch their fa- the most frequent type of failures in the network: 1) opti- vorite movies, but is typically not a high utilization period cal layer recovery (e.g., SONET protection switching [8]), in as far as business applications are concerned. Figure 2 illus- which optical layer failures such as fiber cuts are recovered trates the integrated IP based network architecture. Service rapidly at the optical layer with sub-50ms protection; and 2) routers are connected to backbone routers via dedicated ac- fast re-route (FRR), which sets up an alternate route that cess links (e.g., via a high bandwidth SONET circuit). avoids the failed link [12]. In FRR, as soon as the link fail- SHO ure is detected, all the traffic on the failed link is diverted SHO over a pre-established backup tunnel. At a later time as the IGP and PIM-SSM re-converge, traffic switches back from the backup tunnel to a new optimal path dictated by the routing protocols. Finally, we consider two types of access connections be- Backbone Network VHO VHO tween the service nodes and backbone network nodes: dual- homed and ring. Each node in the network, be it a backbone Existing backbone links for both or a service node, is comprised of two routers to protect IPTV and non-IPTV traffic against a single node failure. In a dual-homed case, each Access links for IPTV Figure 2: IP-based integrated design 3
  4. 4. service node is connected to a backbone node such that the can be optically protected or not – independent of whether two service routers and backbone routers connect in paral- optical layer protection is used on the existing IP infrastruc- lel for fault tolerance. In a ring case, the service nodes in ture. Access links may be dual-homed or interconnected in close geographic proximity to one another are connected as a ring topology. This results in four designs summarized in a ring, where the ring is also connected to the two routers Table 2. in one of the backbone nodes. Figure 3 illustrates the two Table 2: Designs with IP-based dedicated overlay scenarios. Note that there exists a link connecting the two Design Layer Link-cap. Access Fast-failover service routers of a VHO in the ring access. These two ac- Ded-IP-HS IP dedicated dual-homed protected links Ded-IP-HS-FRR IP dedicated dual-homed fast re-route cess mechanisms translate to providing a dual-homed and Ded-IP-Ring IP dedicated ring protected links mesh-based service layer topologies, respectively. Ded-IP-Ring-FRR IP dedicated ring fast re-route Backbone Network Backbone Network 3.1.3 Flat IP Network (No Backbone) We consider constructing a new dedicated IP network purely to carry IPTV traffic. Here, we do not use an IP back- bone network, but rather connect the service routers directly VHO VHO VHO VHO using point-to-point links. High-bandwidth links between VHO VHO Backbone links two service routers can be established directly over dense Access links wavelength division multiplexors (DWDMs) [11]. There- (a) Dual-homed (b) Ring fore, we term this design as “flat IP network” or “flat IP Figure 3: Two types of access connections over DWDM network.” Note that in our case, all the IP designs are built on top of layer 1 optical network. The dis- So far, we have explored the design space in light of tech- tinction between IP and optical designs lies in which layer nology, hierarchy, routing, access connection, and failure multicasting and routing are employed. restoration. This results in four designs which are sum- In designing such a network, we need to identify how the marized in Table 1, where each design is provided a unique service nodes are interconnected. We consider here a meshed name and the technology used. Link-cap. field shows whether topology, where the service nodes carry “through traffic” in the capacity of backbone links are shared with other Inter- addition to acting as sources/sinks of traffic. We use a sim- net applications or dedicated for IPTV traffic, and access ple heuristic to design the topology: we divide the service field shows the type of access connection. Fast-failover field nodes into a small number of communities of interest based shows what type of protection is used at backbone against on geographical proximity. Within a single geographic re- link failures. gion, we connect the service nodes in a ring topology so that Table 1: Designs integrating with IP backbone no single failure can disconnect the network (i.e., traffic can Design Layer Link-cap. Access Fast-failover Int-IP-HS IP shared dual-homed protected links be re-routed around the ring in case of a failure). More Int-IP-HS-FRR IP shared dual-homed fast re-route specifically, the two IP routers within each service node are Int-IP-Ring IP shared ring protected links connected, and also the IP router in one service node is con- Int-IP-Ring-FRR IP shared ring fast re-route nected to other IP router in different service nodes (forming intra- and inter-regional ring connections). Thereby, end- 3.1.2 Dedicated Overlay on Top of an Existing IP to-end paths from SHO to VHO will go through a series of Based Network intermediate service routers. As an alternative IP layer design, we consider a network DWDMs do not traditionally provide multicast capabili- provider overlaying a dedicated topology on top of an ex- ties, and certainly not in the equipment deployed in large isting infrastructure. This is implemented by using com- telecommunications networks. We thus assume that multi- mon backbone routers but dedicated links that support only cast capabilities are not available at the optical layer for this IPTV traffic. The main advantage of this design lies in per- design instance. However, as the service nodes are carrying formance management since links are dedicated for IPTV through traffic, we utilize multicast routing in the service traffic. Figure 4 illustrates the idea, where solid thin and routers themselves. That is, routing is determined by PIM- thick lines represent the backbone links for the existing (non- SSM for multicast and IGP for unicast at service routers. IPTV) traffic and IPTV traffic, respectively. Service nodes Using this design, IPTV traffic is completely isolated from are connected to backbone nodes via dedicated access links. other Internet based traffic that the existing network may carry. Thus, additional performance measures are not re- SHO SHO quired (although, we may still give high priority to IPTV traffic among others). Also, service nodes themselves are directly interconnected, and thus have no access structure. Providing fast failure recovery mechanism is still important in this design, and can be done by providing protection at the optical layer (which translates to using protected links Backbone Network between the service routers) or by relying on FRR in the VHO VHO service routers. This results in two design instances as sum- Existing backbone links marized in Table 3. Backbone links for IPTV Access links for IPTV Table 3: Direct inter-connected optical design Design Layer Link-cap. Access Fast-failover Figure 4: IP-based Design with dedicated overlay P2P-DWDM IP dedicated none protected links The backbone links used for carrying the IPTV traffic P2P-DWDM-FRR IP dedicated none fast re-route 4
  5. 5. 3.2 Optical Based Network Designs signed specifically for carrying only the IPTV traffic. This The conceptually simplest way to connect service routers contrasts with the other network designs which are in no is to provide direct links between them. In contrast with the way optimized for the IPTV traffic as they also support a previous approaches, the paths from SHOs to VHOs are di- wide range of other services. rectly connected by logical links (without traversing through We omit detailed discussion on how each design is real- intermediate IP routers). The link interconnecting is estab- ized due to lack of space and simply note that all design lished over an optical infrastructure, consisting of optical alternatives are realized efficiently to carry the total traffic components such as dense wavelength division multiplexors in the network (both multicast and unicast IPTV traffic, (DWDMs) and optical cross-connects (OXCs) [11]. In the and in the case of the integrated IP design, other non-IPTV following, we consider one optical network alternative. traffic). 1 . In all of the designs, we consistently assume a high level of redundancy, by incorporating two SHOs and 3.2.1 Integration with an Existing Switched Opti- two routers within each service node. We also ensure that cal Network there is sufficient capacity between the service nodes so that We consider a network provider using its switched opti- the design can survive any single failure, including any link, cal network to interconnect the service routers. We assume line card, or router (or optical switch) failure. the optical backbone consists of OXCs. The term here is We compare the capital expenditure across designs. We used somewhat liberally, to cover both all-optical and elec- separately refer to the cost due to the backbone side as the tronic switching. In either case, we assume that the OXC backbone cost and denote the cost of the network-facing ser- can switch any input wavelength/fiber to any output wave- vice routers as the access cost. The cost of a network design length/fiber. If the OXC is electronic, then it can also sup- is represented as the cost of all of the transport cost rela- port time-division multiplexing [8]. Connections between tive to the distance of links and the router/optical switch the service routers are established across one or more OXCs. port cost at the two ends of the links. Parameters for trans- Optical networks defined above have traditionally been port and port cost vary across different technologies (e.g., designed to provide point-to-point connections, such as those port/link vendor/type/capacity). In our evaluation, we as- required to interconnect two IP routers. However, multicas- sume that all the links are OC48 (2.5 Gb/s) and select re- ting capabilities are now being considered for optical net- alistic parameters accordingly. works [13]. Optical networks can multicast their signals by replicating an input signal from a given input port onto two cost 120 IP Network Optical Network or more output ports at an OXC. 100 To effectively make use of the multicast capabilities of- 80 fered by the switched optical network, we assume a hub- 60 multicast unicast and-spoke logical topology for the service node interconnec- 40 tion. Specifically, we assume that each SHO establishes a 20 multicast tree including all the VHOs, simultaneously mul- 0 ticasting the same signal to all VHOs. Each VHO is thereby 1Gbps 2Gbps 3Gbps 1Gbps 2Gbps 3Gbps simultaneously connected to both SHOs. The OXC to which each VHO is connected is then used to select the higher qual- Figure 5: Cost comparison of multicast and unicast ity signal from the two SHOs. We commence by evaluating the effectiveness of multi- Using the above approach, failure recovery is achieved by cast in reducing the network capacity required. Figure 5 rapidly switching between the two signals received by the illustrates the relative cost of carrying broadcast TV traf- OXC directly connected to each VHO. We assume here that fic using multicast versus unicast technologies on both IP the optical network does not re-route internally around a and switched optical networks2 . The Y-axis represents the failure. Thus, the paths from the two SHOs to a common capital expenditure, scaled to hide the proprietary nature of VHO must be physically-diverse from one another to ensure the data. The degree to which multicast benefits depends that traffic can be received even when there is a major fiber on the number of multicast endpoints (the 40 VHOs in this cut. This introduces some additional complexity into the case) and the backbone topology. In general, multicast tech- MDT calculation. In this paper, we use an Integer Pro- nologies will provide more significant gains as the number gramming (IP) based approach to create the trees, ensuring of VHOs increases. In the realistic topologies considered the necessary diversity requirements are met. here, the graph demonstrates that multicasting reduces the Table 4: Designs in optical network backbone network cost by more than a factor of three com- Design Layer Link-Cap. Access Fast-failover pared with unicasting. The reductions in network cost ap- Opt-Switched optical time-divisioned dual-homed disjoint paths pear comparable for both the optical and IP based designs. Next, we compare the capital expenditure across all of 4. EVALUATION OF DESIGNS the designs. Figure 6 shows the relative cost across designs We analyze an IPTV architecture consisting of two SHOs for two different multicast and unicast loads. The cost is and 40 VHO locations across the US. For fair comparison, normalized such that the cost of Opt-Switched for carry- we overlay the service layer on a common backbone net- 1 By “efficiently carrying traffic”, we mean minimizing end-to-end la- work topology for both IP and optical network designs. This tency and the total cost of designs under various failure scenarios. topology is representative of an operational backbone net- 2 For the IP network, both multicast and unicast routing paths are de- work consisting of approximately 100 nodes and 200 links. termined according to Int-IP-HS-FRR design. For the optical network, However, for the flat IP design with no backbone, as de- multicast and unicast paths are determined in a fashion described in Opt-Switched design. Costing of all cases incorporates the extra capac- scribed before, we use a simple heuristic to complete the ity required to service the traffic under each single failure scenarios. design – the outcome being a sparsely connected network de- Using other types of IP designs yield a similar result. 5
  6. 6. Multicast 3Gbps + Unicast 0Gbps Multicast 3Gbps + Unicast 3Gbps 6.0 40.0 5.0 Relative cost 30.0 st 4.0 o c access 3.0 s s ec ca e vi 20.0 t al e no b k c a b backbone e 2.0 R 10.0 1.0 0.0 0.0 P2P-DWDM-FRR P2P-DWDM-FRR Ded-IP-HS-FRR Ded-IP-Ring-FRR Int-IP-HS-FRR Int-IP-Ring-FRR Ded-IP-HS-FRR Ded-IP-Ring-FRR Int-IP-HS-FRR Int-IP-Ring-FRR P2P-DWDM P2P-DWDM Ded-IP-HS Int-IP-HS Ded-IP-Ring Int-IP-Ring Ded-IP-HS Int-IP-HS Ded-IP-Ring Int-IP-Ring Opt-Switched Opt-Switched Figure 6: Cost comparison across design instances ing a multicast load of 1 Gb/s to each VHO is 1.0. Please large amounts of “through” traffic destined for downstream note the different y-axes in the two plots. In this figure, we VHOs, increasing the port and link costs. make the following observations. First, for all of the loads Finally, in terms of the backbone cost, the integrated IP analyzed (including loads not published here), the optical designs are consistently more economical than the dedicated network appeared more economical than the IP network. designs. This illustrates the economic benefit of sharing This is primarily due to the lower port cost of an optical a common infrastructure. However, such benefits must be switch than an IP router – the optical port cost being ap- weighed against the overheads of ensuring isolation of video proximately one third of the cost of an equivalent rate router traffic from the perils of the public Internet. port. In the optical designs, we have assumed an electronic switching fabric that can support time-division multiplex- 5. CONCLUSIONS ing (TDM) and virtual concatenation (VCAT) [4], so that In this paper, we considered a particular problem of sup- we can allocate bandwidth within the backbone in multiples porting IPTV services on a large backbone distribution net- of STS-1 (54 Mb/s) granularity. This allows most effective work. Our work is based on a practical real-world setting use of network capacity, but it is important to note that not and provided what we believe to be the first detailed exam- all technologies deployed today can support this. ination for a single network provider. We focused on one of Second, in IP designs, we observe that the most economi- the critical issues in designing a network to support IPTV, cal approach to achieving fast failover is fast re-route (FRR), namely, capital expenditure. Future work should consider as opposed to using optically protected links. In all of our IP performance trade-offs. Amongst our findings, we demon- designs, we assume that additional capacity is introduced in strated significant benefits of multicast in reducing capital the IP layer to recover from router/port failures, indepen- expense for broadcast TV, illustrated that our particular dent of any other protection mechanism used. This same switched optical network design requires less capital than spare restoration capacity can be used by FRR to provide the IP based designs, and showed that ring access to the rapid failure recovery – thus, FRR only increases the re- backbone is more attractive for broadcast TV, while dual- quired IP layer restoration capacity by a relatively small homed access being more attractive if we have high volume amount compared with that incorporated to handle the sin- of VoD traffic that needs to be unicast in realtime. gle router/port failures. In contrast, the additional capacity introduced at the optical layer to provide rapid optical layer 6. REFERENCES [1] S. Bhattacharyya. An Overview of Source-Specific Multicast protection cannot be shared with the IP layer restoration re- (SSM). RFC 3569, IETF, 2003. quired for handling router failures, and thus effectively dou- [2] R. W. Callon. Use of OSI IS-IS for Routing in TCP/IP and Dual Environments. RFC 1195, IETF, 1990. bles the restoration capacity requirements in the network. [3] S. Casner and S. Deering. First IETF Internet Audiocast. ACM We next compare the access cost of the different solutions. SIGCOMM CCR, 1992. The total network cost is typically dominated by access cost, [4] L. Choy. Virtual Concatenation Tutorial: Enhancing with the exception of the direct DWDM interconnection of SONET/SDH Networks for Data Transport. J. Optical Networking, 2001. VHOs, where access cost is a meaningless concept. In the [5] Cisco. Cisco Gigabit-Ethernet Optimized IPTV/Video over other designs, we observe that ring access is more economical Broadband Solution Design and Implementation Guide. than dual-homed access when carrying only broadcast TV [6] Estrin et al. Protocol Independent Multicast-Sparse Mode (assuming multicast traffic). Connecting neighboring nodes (PIM-SM): Protocol Specification. RFC 2362, IETF, 1998. in a ring topology allows the ring bandwidth to be shared by [7] K. N. et al. Definition of the differentiated services field in the IPv4 and IPv6 headers. RFC 2474, IETF, 1998. all nodes in the ring – making effective use of the multicast- [8] W. Goralski. SONET: A Guide to Synchronous Optical ing capabilities. For N VHOs using ring access, we need to Network. Mcgraw-Hill, 1997. send only a single copy of the (multicasted) broadcast TV [9] Kim et al. Requirements for Internet Media Guides on Internet signals around the ring. In contrast, if each of the N VHOs Protocol Television Services. Draft, IETF, 2005. [10] J. Moy. OSPF Version 2. RFC 2328, IETF, 1998. are dual-homed to the backbone then we need to carry a [11] C. S. R. Murthy and M. Gurusamy. WDM Optical Networks. load proportional to N on the VHO access links. However, Prentice Hall, 2001. when unicast traffic is also incorporated, dual-homed access [12] P. Pan, G. Swallow, and A. Atlas. Fast Reroute Extensions to becomes significantly more cost effective. For large unicast RSVP-TE for LSP Tunnels. RFC 4090, IETF, 2005. [13] Y. Xin and G. N. Rouskas. Multicast Routing Under Optical demands, intermediate nodes in a ring are forced to carry Layer Constraints. In IEEE INFOCOM, 2004. 6